PROJECT SUMMARY/ABSTRACT Relapse remains the major obstacle to cure in leukemia. Why do some patients relapse, while others are cured? The ability to predict response, and intervene with alternative therapies before frank relapse occurs, can potentially optimize therapy, and potentially change the natural history of a particular patient?s disease. We will use AML as a model disease to demonstrate the power of the detection of measurable residual disease (MRD) and gene expression signatures in predicting treatment response. For these studies we will use archival and prospective samples from U.S. Intergroup studies so that we can accurately associate genetic changes with clinical characteristics and outcomes. In Aim 1 we will develop and test a new sequencing method, duplex sequencing, which can detect a mutation at a frequency of one mutant allele in a background of a million wild type alleles. In Aim 2, we will test three gene expression signatures head to head in predicting short and long- term treatment response, and discover pathways different in the response groups that may be future targets of drug therapy. In Aim 3, we will apply the above genetic tests to the allogeneic transplant setting, to predict relapse following non-myeloablative transplantation (and hence, direct abortive therapy). Our lab has had considerable success in using similar tools in CML and ALL, and indeed, MRD detection in these diseases have evolved to a surrogate for outcome in clinical trials. We likewise believe that in AML these aims will eventually allow us to predict outcome prior to therapy, pick therapies based on likely pathways responsible for the predicted unfavorable response, then monitor MRD, changing therapy early if the kinetics of MRD clearance are suboptimal.